SuperAPI AI Directory

artifacts

Frontend & websites

View repository

Updated 10-15 19:55

About this skill

Asta Artifacts

Artifacts are the structured outputs produced by A2A agents. They accumulate in the artifacts[] field of the task response.

Export

# 1. Dump task result
TASK_DIR=$(mktemp -d)
asta <agent> task "<TASK_ID>" > "$TASK_DIR/task.json"

# 2. Export (caller decides output dir and format)
asta artifacts --input "$TASK_DIR" --output "$OUT_DIR" --format md

Produces per-artifact files organized by subtype, plus index.md, references.bib, and manifest.json.

Works on partial task data too — useful for previewing mid-run.

Formats

  • --format md — markdown, one .md per artifact. Cross-artifact links use relative .md paths. Best for LLM consumption, version control, and indexing into asta-documents.
  • --format html — self-contained HTML with index.html. Open with open "$OUT_DIR/index.html". Best for reading or sharing.

Path convention (when invoked by another skill)

When another skill hands off to this one to export a run, default to writing under:

.asta/<invoking-skill>/<slug>/

Where <slug> is YYYY-MM-DD-<short-query-slug> — date-first for chronological ordering, short-slug derived from the primary input (query, topic, etc.). Register the exported artifacts with asta-documents into a single per-skill aggregate index at .asta/<invoking-skill>/index.yaml so consumers can search across all runs.

Example (theorizer run):

.asta/generate-theories/2026-04-17-media-polarization/   # per-run outputs
.asta/generate-theories/index.yaml                        # aggregate index

Indexing with asta-documents

After export, read manifest.json and register each entry with asta documents add. Each manifest entry has: name, description, filename, mime_type, and optionally subtype and metadata. Map these directly to:

  • file://<absolute-path-to-OUT_DIR>/<filename> as the URL
  • --name from name
  • --summary from description (fall back to name)
  • --mime-type from mime_type
  • --tags from subtype if present
  • --extra from metadata as JSON if present

Use --root <dir> to target a scoped index (e.g. per-agent) instead of the default .asta/documents/.